British Journal of Anaesthesia, xxx (xxx): xxx (xxxx)

doi: 10.1016/j.bja.2020.02.027

Advance Access Publication Date: xxx

Clinical Investigation

C L I N I C A L I N V E S T I G A T I O N

Cohort study into the neural correlates of postoperative delirium:the role of connectivity and slow-wave activity

Sean Tanabe1, Rosaleena Mohanty1,2, Heidi Lindroth1,3, Cameron Casey1, Tyler Ballweg1,

Zahra Farahbakhsh1, Bryan Krause1, Vivek Prabhakaran2, Matthew I. Banks1 and

Robert D. Sanders1,4,5,*

1Department of Anesthesiology, University of Wisconsin, Madison, WI, USA, 2Department of Radiology, University of

Wisconsin, Madison, WI, USA, 3Center for Health Innovation and Implementation Science, Center for Aging Research,

Division of Pulmonary and Critical Care Medicine, Regenstrief Institute, Indiana University School of Medicine,

Indianapolis, IN, USA, 4University of Sydney, Australia and 5Department of Anaesthetics, Royal Prince Alfred Hospital,

Australia

*Corresponding author. E-mail: robert.sanders@sydney.edu.au

Abstract

Background: Delirium frequently affects older patients, increasing morbidity and mortality; however, the pathogenesis is

poorly understood. Herein, we tested the cognitive disintegration model, which proposes that a breakdown in fronto-

parietal connectivity, provoked by increased slow-wave activity (SWA), causes delirium.

Methods: We recruited 70 surgical patients to have preoperative and postoperative cognitive testing, EEG, blood bio-

markers, and preoperative MRI. To provide evidence for causality, any putative mechanism had to differentiate on the

diagnosis of delirium; change proportionally to delirium severity; and correlate with a known precipitant for delirium,

inflammation. Analyses were adjusted for multiple corrections (MCs) where appropriate.

Results: In the preoperative period, subjects who subsequently incurred postoperative delirium had higher alpha power,

increased alpha band connectivity (MC P<0.05), but impaired structural connectivity (increased radial diffusivity; MC

P<0.05) on diffusion tensor imaging. These connectivity effects were correlated (r2¼0.491; P¼0.0012). Postoperatively,

local SWA over frontal cortex was insufficient to cause delirium. Rather, delirium was associated with increased SWA

involving occipitoparietal and frontal cortex, with an accompanying breakdown in functional connectivity. Changes in

connectivity correlated with SWA (r2¼0.257; P<0.0001), delirium severity rating (r2¼0.195; P<0.001), interleukin 10

(r2¼0.152; P¼0.008), and monocyte chemoattractant protein 1 (r2¼0.253; P<0.001).Conclusions: Whilst frontal SWA occurs in all postoperative patients, delirium results when SWA progresses to involve

posterior brain regions, with an associated reduction in connectivity in most subjects. Modifying SWA and connectivity

may offer a novel therapeutic approach for delirium.

Clinical trial registration: NCT03124303, NCT02926417

Keywords: cognitive dysfunction; connectivity; delirium; electroencephalogram; inflammation; mechanism; post-

operative; slow wave activity; surgery

Received: 4 November 2019; Accepted: 28 February 2020

© 2020 British Journal of Anaesthesia. Published by Elsevier Ltd. All rights reserved.

For Permissions, please email: permissions@elsevier.com

1

Editor’s key points

� Electroencephalographic slow-wave activity has previ-

ously been associated with delirium.

� We found local SWA in patients without delirium but

global SWA in patients with delirium.

� A distinguishing feature of postoperative delirium in

most patients was a decrease in correlates of fronto-

parietal connectivity.

� Decreased preoperative structural rather than func-

tional connectivity was predictive of postoperative

delirium.

2 - Tanabe et al.

Delirium is a sudden onset of confusion that frequently affects

older patients.1 It is associated with increased morbidity,

mortality, loss of functional independence, and worsening

quality of life, and may herald the onset of cognitive decline

and dementia.1,2 It is estimated to cost $160 billion per annum

in the USA.3 The pathogenesis of delirium remains obscure;

whilst several theories have been proposed,4e7 no therapies

have been forthcoming. We proposed a cognitive disintegra-

tion model,7 whereby the severe cognitive symptoms of

delirium are related to breakdown in connectivity driven by

increased EEG slow-wave activity (SWA) in key frontoparietal

cognitive networks. Inspired by the basic tenets of cognitive

frameworks, such as integrated information theory8 and pre-

dictive coding,9 we surmised that cognition requires the inte-

gration of information from functionally specialised brain

regions to form a coherent view of self and the environment.

Delirium is characterised by fragmented perceptions and

disjointed interactions with the environment, consistent with

disintegration of certain cognitive processes. Preliminary evi-

dence for this model, based on reductions in connectivity

during delirium, has been provided by Numan and col-

leagues10 and van Dellen and colleagues.11 However, the case-

control design used does not exclude premorbid differences

between the groups. We hypothesised that pre-existing de-

creases in connectivity would predispose to delirium.7 Thus,

we collected EEG data preoperatively, before delirium

occurred. The electrophysiological hallmark of delirium is

cortical SWA. Cortical SWA is a characteristic of both deep

non-rapid eye movement sleep12 and anaesthesia.13 In both of

these states, it is associated with reduced connectivity. Hence,

it is plausible that delirium-associated SWA may lead to im-

pairments in connectivity.

As there is presently no way of reversing delirium to

establish causality, we used the Bradford Hill criteria for

causation to support the importance of the mechanisms un-

covered. Hence, we sought a strong association, with specific

changes in the delirious population that changed contempora-

neously with delirium and with a biological gradient. To achieve

these criteria, the change with transition to delirium must

differ from non-delirious subjects and should also change

proportionately to delirium symptom severity (henceforth,

delirium severity). These criteria also impose a relatively

stringent statistical threshold: requiring a test to be significant

for both delirium incidence and severity would exclude

spurious results at a threshold equivalent to 0.052. Further-

more, to provide biological plausibility, we hypothesised that

relevant effects should also correlate with inflammation, a

probable precipitant of delirium. To identify relevant markers

of systemic inflammation, we identified plasma cytokines that

changed with both delirium incidence and severity.

Whilst multiple studies have shown that delirium is asso-

ciated with diffuse slowing in the EEG,14 they typically have

not accounted for pre-delirium status or within- and across-

subject differences. Nor have they looked for Bradford Hill

criteria of causation nor related findings to severity or

inflammation. Herein, we took advantage of major elective

surgery, where the precipitating event for delirium is pre-

dictable in time. We tested our hypothesis that, when SWA

progressed to involve frontoparietal brain regions, the asso-

ciated breakdown in connectivity would precipitate delirium.7

Methods

The data are derived from two ongoing perioperative cohort

studies registeredwith https://clinicaltrials.gov/(Interventions

for Postoperative Delirium: Biomarker-3 (IPOD-B3),

NCT03124303, and Interventions for Postoperative Delirium:

Biomarker-2 (IPOD-B2), NCT02926417) and approved by the

University of WisconsineMadison Institutional Review Board

(#2015-0374; 2015-0960). EEG data were collected from 72 pa-

tients before and after surgery between 2015 and 2019. Two

subjects’ data were excluded because of poor impedance

contaminating the entire recording. After exclusion of these

two subjects, all available EEG data (in subjects with paired

preoperative and postoperative EEG) collected until 70 subjects

were recruitedwere used in the analyses (Fig. 1). Eight subjects

were recruited only to IPOD-B2 and 62 to IPOD-B3. Sensitivity

analyses, conducted only on subjects recruited to IPOD-B3,

found similar results. IPOD-B2 is a small ongoing cohort

study of adult subjects undergoing thoracic aortic aneurysm

repair (open or endovascular; Supplementary Table 1). IPOD-

B3 is a larger ongoing cohort of adult patients (>65 yr old)

undergoing major surgery (Supplementary Table 1).

Cohort sample size

A priori power analysis suggested that a minimum of 59 sub-

jects would be required to show EEG differences in delirium.

Prior data suggested an effect size for a ‘between-subject’

analysis of a three-fold increase in SWA11 in delirium. Based

on the variance in SWA from van Dellen and colleagues,11

power analysis suggests that 12 delirious subjects, compared

with 47 controls, will provide 99% power for an uncorrected

a<0.001 to detect SWA differences assuming a 20% delirium

incidence. Calculation of SWA differences required paired EEG

samples for preoperative and postoperative data from which

we could calculate the change in the delirious subjects and

compare with the change in the non-delirious subjects.

Delirium diagnosis

The 70 subjects were grouped depending on postoperative

confusion assessment method diagnoses of delirium (confu-

sion assessment method [CAM]þ/e15,16 or CAM-ICUþ/e if

ventilated17) status at the time of the EEG recording as delir-

ious or not.15 The delirium group contained 22 subjects and the

non-delirium group contained 48 subjects. Delirium severity

was scored using the Delirium Rating of Severity-98 (DRS).18

Weekly delirium review meetings were held to ensure con-

sistency of scoring during the data collection period.

Fig 1. Study design and Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) diagram. (a) Study design.

Delirium severity (DRS) is collected during both preoperative and postoperative delirium assessment. (b) STROBE diagram. Data are from

IPOD-B2 and IPOD-B3 ongoing perioperative cohort studies. DRS, Delirium Rating of Severity-98; DTI, diffusion tensor imaging; IPOD-B2,

Interventions for Postoperative Delirium: Biomarker-2; IPOD-B3, Interventions for Postoperative Delirium: Biomarker-3.

Neural correlates of delirium - 3

EEG recording and preprocessing

EEG data were collected using high-density EEG with 256 chan-

nels (Electrical Geodesics, Inc., Eugene, OR, USA) using a sam-

pling frequency of 250 Hz (five subjectswere recorded at 500 Hz;

these were down sampled to 250 Hz). Data were collected pre-

operativelywithamedianof5daysbeforesurgery (inter-quartile

range [IQR]: 1e7 days) and postoperatively with amedian of the

first postoperative day (IQR: 1e1 day). For each subject and

condition, 15 min of eyes-closed resting-state data were

collected. Inthreesubjects, eyes-opendata foraminimumof90s

were collected allowing comparison of eyes open and closed.

These data were filtered from 0.1 to 50 Hz using Hamming

windowed-sinc FIR filter. Channels and data segments con-

taining artifact were visually inspected and removed. To

remove artifact of eye movement and muscle movement, we

applied independent component analysis in EEGLAB. Before

analysis, the removed channels were interpolated, and then

the signals were average-referenced.

4 - Tanabe et al.

Spectral power

The data were estimated for power spectral density with

Welch’s method into delta (0.5e4 Hz), theta (4e6 Hz), alpha

(6e12 Hz), beta (12e28 Hz), and gamma (28e40 Hz). The power

spectral density was log10 transformed to normalise the dis-

tribution. Slow-wave activity was post hoc defined as 0.5e6 Hz

similar to prior definitions.19

Phase lag index

To measure the functional connectivity between channel sig-

nals, the debiased weighted phase lag index (PLI) in the

FieldTrip toolbox (maintained by the Donders Institute for

Brain, Cognition, and Behaviour which is part of Radboud

University in Nijmegen, the Netherlands) was used.20 Phase

lag index alone estimates the degree to which the two time-

series leads and lags by indexing their phase differences.

However, PLI is sensitive to signal synchronisations because of

volume conduction. By weighting the phases across cross-

spectrum time series, PLI is able to mitigate the sensitivity to

volume conduction.20 Debiasedweighted PLI further estimates

the squared weighted PLI to normalise the positive bias.

Between all 256 channels, in a pairwise manner, we

calculate debiased PLI using 2 s window length and 25%

overlap. To probe changes associated with detectable SWA in

both the delirious and non-delirious groups, we divided the

256 scalp electrodes in 25 squares (Supplementary Fig. 1). The

PLI between these ‘squares’ were averaged, producing

pairwise-squared PLI. We focused on PLI involving the midline

frontal square centred on the Fz electrode, which showed

increased SWA in both delirious and non-delirious subjects.

However, we also conducted pairwise analyses across con-

nections and correlated these changes in connectivity with

delirium severity. These analyses were corrected by false

discovery rate (FDR) methods. We also plot a PLI spectrum by

using the cross-spectra produced from multi-taper frequency

transformation (see Supplementary data for methods).

EEG statistical analysis

To test for topographical differences or correlations, two-

tailed statistical non-parametric mapping (SNPM) threshold-

free cluster enhancement was used.21 Wilcoxon rank-sum

test, Wilcoxon signed-rank test, and Pearson correlation

were used for non-topographical tests where appropriate. To

test for differences amongst groups (across delirium status

and pre-/postoperation), we used linear mixed-effects models,

testing the interaction of delirium status and time point (pre vs

post) for each frequency band. Test statistics and P-values

were calculated using Type III Wald F-tests with

KenwardeRoger degrees of freedom.

Topographical PLI analyses use parametric Pearson corre-

lation on each PLI connection. These analyses are sensitive to

outlier influence, as parametric tests assume a homogeneous

variance. Therefore, for each PLI connectivity, we excluded

outlier subjects whose Cook’s distance is above 4*(mean of

Cook’s distance).22 Excluding subjects with large Cook’s dis-

tances resulted in substantial changes to the regressionmodel

estimation. As each PLI connectivity topoplot has a different

regression model, there will be a different number of subjects

excluded for each connection. Because of this, analyses mak-

ing use of Cook’s distance will indicate sample size as a range

(e.g. n¼70 [64e69, after outlier exclusion]). The PLI analyses

were corrected by FDR methods.

Blood biomarkers

We also collected blood samples for biomarker analyses

immediately before and after the resting-state EEG. Plasma

samples were collected in ethylenediamine tetra-acetic acid

(EDTA)-containing tubes preoperatively and stored at e80�C.

Cytokine analysis

A multiplex assay was conducted at Eve Technologies (Mon-

treal, Canada) for plasma samples that were paired with EEG

collection. The cytokines measured were interleukin (IL) 1 (IL-

1) beta, IL-1 receptor antagonist, IL-2, IL-4, IL-6, IL-8, IL-10, IL-

12, monocyte chemoattractant protein 1 (MCP-1), and tumour

necrosis factor-alpha. For the cytokine analysis only, subjects

were excluded if they had no blood drawn around the time of

the EEG recording.

Diffusion tensor imaging methods

A post hoc decisionwasmade during the analysis of EEG data to

investigate preoperative structural connectivity (diffusion

tensor imaging [DTI]) in patients with EEG to investigate an

unexpected findingwith preoperative frontal EEG connectivity.

All availableDTI data in the cohort are reported. Of the patients

recruited to IPOD-B3, 75 patients consented to preoperative

imaging (there is no imaging collected in IPOD-B2). Six of the

DTI scans were discarded because of motion artifact. Preoper-

ative DTI scans were available from 68 patients (17 delirious)

after quality control and preprocessing. Thirty-six subjects had

preoperative imaging and EEG.MRI datawere collected on 3.0 T

whole-body scanner (General Electric Medical Systems, Wau-

kesha, WI, USA) with an eight-channel head coil. Diffusion

tensor imaging datawere acquiredwith the followingdiffusion

parameters: single-shot echo planar imaging, repetition

time¼9000ms, echo time¼66.2ms, single average (NEX) 1, field

of view 256�256 mm2, voxel size 1�1�2 mm3, 75 axial slices

with no gap between slices, flip angle 90�, 56 gradient encoded

directions, and b-values of 0 and 1000 s mm�2.

The raw DTI were checked manually for quality control,

and each remaining scan was corrected for head motion and

eddy-current distortionwith the eddy23 tool and skull stripping

and removal of non-tissue components in FSL.24 The diffusion

tensor model, based on the least-squares fit, was applied using

FDT’s dtifit providing data on fractional anisotropy (FA), mean

diffusivity (MD), axial diffusivity (AD), and radial diffusivity

(RD). Non-linear registration was conducted using Advanced

Normalization Tools (ANTs is maintained by the Penn Image

Computing and Science Lab which is operated out of the

University of Pennsylvania in Philadelphia, PA, USA)25 to a

standard template based on the publicly available Mayo Clinic

Adult Lifespan Template atlas.26

Prior data emphasised RD,27 and hence, we pursued that as

our primary DTI metric, although convergent results were

obtained with MD. Voxel-wise skeletonised white matter tract

maps were obtained for each DTI metric with tract-based

spatial statistics (TBSS),28 applied to AD, MD, and RD, and

were used to perform cross-subject statistical analyses. Voxel-

wise statistical approach (randomise tool21 with threshold-free

cluster enhancement) was adopted to study the association

between the preprocessed and TBSS-based white matter

Fig 2. Delirium occurs with increased global slow-wave activity and loss of posterior high-frequency activity. (a) Representative EEG signals

from two electrodes, one frontal (Fz) and one posterior/occipital (Oz), collected preoperatively (Preop/Pre) and postoperatively (Post). (b)

Group-level spectral analysis comparing patients preoperatively and postoperatively for patients who do (D), or do not (ND), incur delirium

(P<0.05 [black squares], linear mixed-effects model across delirium status and operative phase). Spectra range from 0.5 to 40 Hz, statistics

were completed for frequency bands; d 0.5e4 Hz, q 4e6 Hz, a 6e12 Hz, b 12e28 Hz, g 28e40 Hz. The alpha band range (a), 6e12 Hz, was

chosen by inspecting the peak in the mean spectra amongst all subjects. Topological spectral power of the preoperative and postoperative

states for patients who do not incur delirium (c) or do incur delirium (d) across each power band. The third row shows the post-

operative>preoperative contrast t-maps, corrected with threshold-free cluster enhancement (TFCE), with overlaid electrodes showing

significant differences (statistical non-parametric mapping [SNPM] paired; corrected TFCE P<0.05 [white dots]). (e) Spectral analysis for the

transition in state (postoperatively-preoperatively) for subjects who incur delirium or do not (P<0.05 [black squares] after FDR correction

with two-sided rank-sum test) for electrodes Fz and Oz. Grey square: uncorrected P<0.05. (f) Topological contrasts of the transition to the

non-delirious vs delirious state (first two rows). Third row shows the TFCE t-map of two groups contrast with significant electrodes

overlaid (SNPM unpaired; TFCE corrected P<0.05 t-map). For all statistics, nND¼48 and nD¼22. PSD, power spectral density.

Neural correlates of delirium - 5

Fig 3. Posterior slow-wave activity (SWA) is associated with (a) delirium severity and (b, c) systemic inflammation. Each column represents

the correlation of change in power (postoperatively-preoperatively) with an outcome either Delirium Rating of Severity-98 (DRS), inter-

leukin (IL)-10, or monocyte chemoattractant protein 1 (MCP-1). Top row shows correlation with transition in spectral power across fre-

quencies from occipital electrode (Oz). Spectra range from 0.5 to 40 Hz; statistics done on frequency bands; d 0.5e4 Hz, q 4e6 Hz, a 6e12 Hz,

b 12e28 Hz, and g 28e40 Hz. Black square: P<0.05 after false discovery rate correction. Grey square: P<0.05 uncorrected. Middle row shows

topological correlation with the increase in SWA (0.5e6 Hz; statistical non-parametric mapping correlation; corrected threshold-free

cluster enhancement P<0.05 [white dots show statistically significant electrodes]). Bottom row shows correlation with transition in

SWA from occipital electrode Oz. DRS n¼70; IL-8 n¼47; IL-10 n¼46; MCP-1 n¼47.

6 - Tanabe et al.

skeletonised maps and delirium incidence, and severity on a

group level (using 10 000 permutations). Voxel-wise statistical

significance inferences were drawn by controlling the family-

wise error (FWE) rate at a 0.05 threshold. Detailed descriptions

of the DTI methods are available in the Supplementary

material.

Fig 4. Delirium is associated with decreased connectivity (debiased weighted phase lag index [PLI]). (a) PLI spectral analysis comparing

patients preoperatively (pre) and postoperatively (post) for patients who do (D), or do not (ND), incur delirium (P<0.05 [black squares]).

Statistical analysis using a linear mixed-effects model across delirium states and pre-/postoperation is reported (nND¼48 and nD¼22).

Spectra range from 0.5 to 40 Hz; statistics done on frequency bands; d 0.5e4 Hz, q 4e6 Hz, a 6e12 Hz, b 12e28 Hz, and g 28e40 Hz. (b)

Topologically PLI of preoperative and postoperative states for patients who do not, or do, incur delirium. PLI with a frontal square group of

electrodes is displayed and corrected. Bottom row shows postoperative>preoperative contrast (P<0.05 [line displayed if significant] after

false discovery rate (FDR) correction with two-sided Wilcoxon signed-rank test). Right column shows contrast between patients who do,

and do not, incur delirium, delirium>no delirium (P<0.05 after FDR correction with two-sided Wilcoxon rank-sum test; nND¼48 and nD¼22).

(c) Topological plots of the change in connectivity with transition to the non-delirious vs delirious state. Right, topoplot shows threshold-

free cluster enhancement t-map of two groups contrast (P<0.05 after FDR correction with two-sided Wilcoxon rank-sum test; nND¼48 and

nD¼22). (d) Correlation of the postoperativeepreoperative difference in PLI with changes in Delirium Rating of Severity-98 (DRS), inter-

leukin (IL)-10, monocyte chemoattractant protein 1 (MCP-1), and slow-wave activity (SWA). Left column shows r2 map (P<0.05 after FDR

correction; DRS n¼70 [64e69, after outlier exclusion], IL-10 n¼46 [42e45], MCP-1 n¼47 [42e46], and SWA n¼70 [63e68]). Right columns show

correlations with PLI of frontal square connecting to square below.

Neural correlates of delirium - 7

Fig 5. Preoperative connectivity is increased before delirium and inversely correlates with structural connectivity. (a) Preoperative radial

diffusivity (RD) contrast for patients who do not incur delirium (ND) and do incur delirium (D) (general linear model with covariate

ageþgender; P<0.05 family-wise error corrected; nND¼51 and nD¼17). Significant effects of RD (red) are overlaid on whole brain white

matter tract (green). The right box plot shows mean RD contrast across significant voxels detected with multiple comparison correction

between the groups (P<0.05 corrected; two-sided Wilcoxon rank-sum test). (b) Correlation of preoperative RD with peak postoperative

delirium severity (general linear model with covariate ageþgender; P<0.05 FWE corrected; n¼68). (c) Correlation of preoperative (preop)

topological alpha phase lag index (PLI) and mean RD across significant voxels (P<0.05 after false discovery rate correction; n¼36 [33e36

after outlier exclusion]). Left, topoplot shows r2 map for the PLI of frontal square is displayed and corrected. Right plot shows mean RD

correlation with PLI of frontal square connecting to parietal square. DRS, Delirium Rating of Severity-98.

8 - Tanabe et al.

Neural correlates of delirium - 9

Results

Figure 1 shows the Strengthening the Reporting of Observa-

tional Studies in Epidemiology (STROBE) diagram and data

collection strategy. Cohort characteristics are described in

Supplementary Table 2. The median postoperative DRS was 9.

Postoperative delirium occurred in 31% of patients, whose

median DRS was 17. Delirious patients had similar age and sex

to patients who did not become delirious, but they underwent

operations with higher surgical severity (Wilcoxon rank-sum

test; P<0.001; Supplementary Table 2).

EEG power spectral analyses

Initially, we focused on the within-subject change in the EEG

that occurs from preoperative to postoperative phases of care.

This showed obvious slowing on transition to the delirious

state (Fig. 2a). For quantitative analysis, the EEG was initially

segmented into different power bands, delta (0.5e4 Hz), theta

(4e6 Hz), alpha (6e12 Hz), beta (12e28 Hz), and gamma (28e40

Hz), based on our division of data around the alpha peak that is

evident in Fig. 2b. Our data show an increase in delta and theta

power in frontal regions in both the non-delirious (linear

mixed-effects model across delirium states and pre-/post-

operation Pdelta<0.0001 and Ptheta<0.0001; Fig. 2c) and delirious

(Pdelta<0.0001 and Ptheta<0.0001; Fig. 2d) subjects. A critical dif-

ference between patient groupswas that delta and theta power

increased in posterior regions in delirium, with a reciprocal

decrease in higher-frequency activity (Fig. 2b and d). Interest-

ingly, patients who later incurred delirium had higher frontal

alpha power preoperatively (P¼0.0338; Fig. 2b), but no change

postoperatively. In contrast, frontal alpha power increased

postoperatively in patients who did not incur delirium (Fig. 2c).

The preoperative differences between the subjects who did

not incur delirium postoperatively and those that did

emphasised the need to focus on the differences in EEG dy-

namics that occur on transition to delirium. Hence, we next

confirmed that the changes in SWA that occur on transition

from preoperative to postoperative state differed between the

groups (described in Supplementary Fig. 2). This analysis was

conducted to exclude non-specific EEG changes that may

occur postoperatively (such as small increases in frontal SWA).

In this focused analysis, increased global delta and theta po-

wer and loss of posterior high-frequency activity appeared

characteristic of delirium (Fig. 2e and f).

Comparison of eyes-open and eyes-closed powerspectra

The global increase in delta and theta power in delirious pa-

tients was surprising given that the subjects were sufficiently

wakeful to complete our cognitive assessments. To confirm

that the subjects were wakeful, we compared the power

spectra for delirious subjects recorded with eyes open and

closed (the latter being the typical state during the recordings).

Whilst only three delirious patients were captured with eyes

open (as delirious patients struggle to follow commands),

there was no difference in the power spectra between the

eyes-open and eyes-closed states, suggesting we were not

merely studying sleep (Supplementary Fig. 3).

EEG power spectra sensitivity analyses

We next confirmed that this effect was not driven by patients

requiring sedation for intubation on the ICU by performing a

sensitivity analysis, excluding these seven patients

(Supplementary Fig. 4). The results were unaffected by

excluding these seven patients. As delta and theta bands

showed similar effects across all these contrasts, henceforth

we treated them as one frequency range, SWA, as has been

done in studies of sleep deprivation.19,29

Correlations of EEG power and delirium severity

Next, we hypothesised that, as SWA increased, delirium

severity would worsen. We calculated the Pearson correlation

of power, for each frequency band, with the change in

delirium severity (DRS). For the purpose of display, we also

show the correlation coefficient in 0.1 Hz frequency bins.

Slow-wave activity showed a remarkable correlation with DRS

with specificity at <6 Hz, such that, as SWA increased,

delirium severity increased (Fig. 3a). Topographic analysis

showed that, whilst there were significant correlations of DRS

and SWA inmany electrodes across the scalp, the relationship

appeared strongest in posterior regions. Indeed, at electrode

Oz, the correlation was relatively strong (r2¼0.207; P¼0.0001).

However, there was no statistical difference between Fz and

Oz for the correlation with DRS with a Williams test; P¼0.67.

We identified a similar effect when excluding the seven pa-

tients on sedation (Supplementary Fig. 4), and confirmed that

all delirious subjects showed increases in Oz SWA

(Supplementary Fig. 5). Correlations of DRS with EEG changes

in higher frequencies were not observed, suggesting these

changes may not be causally related to the delirium state

(Supplementary Fig. 6). These data suggest that SWA is a key

feature of delirium.

Correlations of EEG power and cytokines

We looked for associations between cytokine changes and

delirium using a multiplex panel of 10 cytokines (Eve Tech-

nologies) using Bonferroni multiple comparison correction.

Consistent with prior studies,30 we identified significant asso-

ciations of simultaneous changes in MCP-1 (r2¼0.192; P¼0.002)

and IL-10 (r2¼0.200; P¼0.002), with both delirium severity (DRS;

Supplementary Fig. 7) and delirium incidence (Supplementary

Fig. 8). The other cytokines did not achieve significance on

both tests. Given the detected and plausible association with

delirium, we tested whether MCP-1 and IL-10 were also corre-

lated with SWA. Similar <6 Hz changes in posterior electrode

SWA were correlated with these two cytokines (Fig. 3c and d),

suggesting that systemic inflammation may drive SWA.

EEG connectivity analyses

To address whether integration also fades in delirium, we

measured functional connectivity using PLI.20 We hypoth-

esised that, whilst local SWA in frontal cortex was insufficient

to cause delirium, when SWA transitioned to be global, the

ensuing reduction in cortical connectivity precipitated

delirium. First, we confirmed that alpha band was the appro-

priate frequency range to explore in our data (as suggested by a

prior study11) by plotting the PLI spectrum in 0.4883 Hz fre-

quency resolution. Indeed, there was a clear peak in connec-

tivity in the alpha range preoperatively (linear mixed-effects

model across delirium states and pre-/postoperation Pal-

pha<0.0001; Fig. 4a). Delirium was also associated with an in-

crease in Fz-Pz connectivity in the delta range (Fig. 4a).

10 - Tanabe et al.

To make our analysis sensitive to change in the non-

delirious group (where we observed increases in SWA post-

operatively), we placed a connectivity seed in frontal regions

(see Supplementary Fig. 1 for methodological description) and

calculated connectivity with other scalp regions. In case-

control analyses (Fig. 4b rows), preoperatively, patients who

subsequently became delirious showed increased frontal

connectivity. Postoperatively, however, they showed

decreased connectivity.

Change in connectivity from preoperative levels

We next analysed the change in connectivity from preoperative

to postoperative state across groups. In the non-delirious

subjects, there was no change in connectivity from the pre-

operative period, consistent with only localised SWA (Fig. 4b,

columns). However, the transition into delirium was associ-

ated with significant decreases in frontal connectivity with

posterior brain regions (Fig. 4c). Of delirious subjects, 19

showed decreases in Fz-Pz alpha PLI and three showed in-

creases (Supplementary Fig. 9).

Connectivity sensitivity analyses

The results without outlier rejection are presented in

Supplementary Fig. 10. We confirmed changes of connectivity

that correlated with DRS with calculations of pairwise con-

nectivity between all electrodes and a conservative FDR

correction across pairwise connections (Supplementary

Fig. 11). These data emphasised the correlations of fronto-

temporal connections with delirium severity. In sum, our data

suggest that, in delirium, as SWA extends to reach posterior

brain regions, there is an associated breakdown in brain

connectivity.

Possible mechanistic associations of the change inconnectivity

Changes in connectivity were also correlated with SWA

(r2¼0.257; P<0.0001), DRS (r2¼0.195; P<0.001), IL-10 (r2¼0.152;

P¼0.008), and MCP-1 (r2¼0.253; P<0.001; Fig. 4d).

Post hoc investigation into preoperative structuralconnectivity and delirium

One aspect of these connectivity observations is particularly at

odds with our published hypothesis of delirium.7 Preopera-

tively, subjects who would later become delirious had higher

EEG functional connectivity than subjects who did not become

delirious. Prima facie, this observation also seemed to contra-

dict extensive prior data showing impaired structural con-

nectivity before delirium; see for example Cavallari and

colleagues.27 Therefore, we analysed available structural

connectivity using preoperative DTI. Consistent with prior

data,27 preoperatively, patients before delirium had impaired

structural connectivity, reflected by increased RD, adjusted for

age and gender as detected using TBSS in FSL (voxel-wise

statistics: number of voxels¼4260, FWE P¼0.047, t¼4.5; mean

statistics: RDdelirium>RDnon-delirium with Wilcoxon rank-sum

test; P<0.001; Fig. 5a; Supplementary Table 3). Again, similar

to prior results,27 increased preoperative RD values also

correlated with increasing postoperative delirium severity,

adjusted for age and gender (voxel-wise statistics: number of

voxels¼5595, FWE P¼0.044, t¼5.23; mean statistics: association

of mean RD in significant voxels and delirium severity was

r2¼0.29, P<0.001; Fig. 5b). Similar findings were observed in MD

as well, when adjusted for age and gender (delirium incidence:

voxel-wise statistics: number of voxels¼4757, FWE P¼0.026,

t¼5.33; mean statistics: mean statistics: MDdelirium>MDnon-

delirium with Wilcoxon rank-sum test, P<0.001; delirium severity:

voxel-wise statistics: number of voxels¼3747, FWE P¼0.036,

t¼4.41; mean statistics: association of mean MD in significant

voxels and delirium severity was r2¼0.2, P<0.001;Supplementary Fig. 12; Supplementary Table 4). Associations

with AD and FA were not identified (data not shown).

Post hoc investigation into preoperative structuralconnectivity and EEG functional connectivity

We hypothesised that increased functional connectivity in the

EEG may represent a mechanism of compensation for under-

lying structural pathology based on recent hypotheses from

neuroimaging.31 Next, we quantified the mean value of RD

across voxels that correlated with delirium severity, and

looked for how these correlated with the preoperative EEG

connectivity measures. To do this, we correlated the measure

with topographic EEG connectivity, testing whether the func-

tional connectivity changes in the EEG reflected a compensa-

tory response to the underlying structural pathology. Positive

correlations were noted, focused on left frontal regions, for the

relationship of RD and alpha band frontal connectivity (elec-

trode Fz r2¼0.491; P¼0.0012; Fig. 5c). These data suggest that, as

structural connectivity worsened (higher RD) on the right,

functional connectivity of the EEG increased on the left, sup-

porting the concept that increased preoperative functional

connectivity is a compensatory mechanism for underlying

structural neurodegeneration.

Discussion

Our data suggest that local frontal SWA can occur post-

operatively, but does not result in delirium. Rather, post-

operative delirium results when inflammation-driven SWA

extends to involve occipitoparietal cortex associated with a

breakdown in connectivity. Our prior hypothesis7 emphasised

the role of connectivity in delirium; a subsequent study sup-

ported this idea,11 but did not relate the changes mechanisti-

cally to SWA, inflammation, or investigate proportionality to

delirium severity. That study was also limited by its case-

control design. We have shown clear links to inflammatory

models of delirium through correlation of contemporaneous

changes in systemic cytokines, and relate the connectivity

changes to SWA. Whilst we are presently unable to manipu-

late delirium experimentally to test causal mechanisms, our

data are strongly supported by several Bradford Hill criteria for

causality: temporal sequence, strength of change, and dose

dependence. Our data also show coherence with the dementia

literature32e36 and biological plausibility based on both the links

between inflammation and delirium,37 and also the changes in

SWA following sleep deprivation and consequent impair-

ments in cognition.19,29

An unexpected finding was that of increased frontal high-

frequency power and alpha band connectivity before

delirium. We speculate that these changes were compensa-

tory responses to underlying neurodegeneration, and provided

evidence for this from structural imaging data. Increased

frontal alpha power has recently been identified as a signature

of amyloid disease.38 As an extension of recent hypotheses

Neural correlates of delirium - 11

from neuroimaging,31 we surmise that these electrophysio-

logical changes may represent evidence of functional com-

pensations whereby increased network connectivity is

recruited as a mechanism tomaintain cognitive function from

chronic neurodegenerative pathologies. Interestingly, the

non-delirious subjects showed an increase in frontal alpha

power and maintained cognitive function postoperatively.

Increased alpha activity and connectivity may be a key

compensatory mechanism to maintain cognitive function in

the presence of a physiological stressor (such as inflamma-

tion). Increasing 8 Hz frontotemporal connectivity (within our

alpha range) has recently been shown to improve cognitive

performance in the older population.39We have replicated and

extended prior work10,11 showing that decreases in alpha band

connectivity may contribute to the cognitive dysfunction in

delirium, although it is important to note that decreases in

alpha band connectivity did not occur in all delirious subjects.

It is presently unclear whether these subjects showed con-

nectivity changes in other bands leading to delirium, or

whether connectivity changes do not contribute to delirium in

these subjects. Nonetheless, manipulation of alpha band

connectivity may yet prove to be a therapy for delirium in at

least some subjects.

We note some important limitations of our study. Firstly,

the cohort was of modest size, and it may be that more diverse

changes may be identified in a larger cohort, or alternative

changes may occur in other types of delirium, for example,

those induced by medications. We acknowledge these limita-

tions, but stress that we have likely identified the strongest

relationships, which are most likely to be causal, and that

inflammation is a very common mechanism of delirium6 (as

delirium by definition arises from a physiological stressor).

Nonetheless, our focus on inflammation does not imply that

the associations we have discovered will extend to other

clinical settings or mechanistic types of delirium, where pa-

thologies other than inflammation may play a larger role.

Secondly, it is not possible to perform more invasive electro-

physiological studies in most types of delirium. Thirdly, it is

possible that some form of unmeasured confounding may still

affect our results; however, we control for the baseline state

and calculated changes in the perioperative period. This

approach, essentially normalising to the baseline state, ad-

justs for many of the confounds of the baseline state,

including age and co-morbidities. This is critical, given we

have observed preoperative differences in the EEG. We also

conducted sensitivity analyses for cohort type (IPOD-B2 vs

IPOD-B3) and exposure to sedative medications.

In summary, our data show that changes in occipitoparietal

cortical SWA correlated with worsening delirium severity,

systemic inflammation, and impaired connectivity. Hence,

whilst frontal, local SWA occurs in postoperative patients,

delirium results when SWA progresses to involve posterior

brain regions, with an associated reduction in connectivity.

Authors’ contributions

Study design: RDS, MIB

Data collection/analysis: RDS, ST, HL, CC, ZF, TB, RM

Devising of imaging protocol: VP

Imaging analyses: RM, VP

EEG analyses: ST, RDS, MIB

Statistical advice: BK

Writing of paper: RDS, ST

Declaration of interest

All authors declare no competing interests that may be rele-

vant to the submitted work.

Funding

National Institutes of Health (K23 AG055700) to RDS; National

Institutes of Health (R01 AG063849-01) to RDS; National Heart,

Lung, and Blood Institute (5T32HL091816-07) to HL.

Acknowledgements

The authors thank Drs Brady Riedner and Giulio Tononi for

advice and loan of EEG equipment.

Appendix A. Supplementary data

Supplementary data to this article can be found online at

https://doi.org/10.1016/j.bja.2020.02.027.

References

1. Inouye SK, Westendorp RG, Saczynski JS. Delirium in

elderly people. Lancet 2014; 383: 911e22

2. Witlox J, Eurelings LS, de Jonghe JF, Kalisvaart KJ,

Eikelenboom P, van Gool WA. Delirium in elderly patients

and the risk of postdischarge mortality, institutionaliza-

tion, and dementia: a meta-analysis. JAMA 2010; 304:

443e51

3. Leslie DL, Marcantonio ER, Zhang Y, Leo-Summers L,

Inouye SK. One-year health care costs associated with

delirium in the elderly population. Arch Intern Med 2008;

168: 27e32

4. Hshieh TT, Fong TG, Marcantonio ER, Inouye SK. Cholin-

ergic deficiency hypothesis in delirium: a synthesis of

current evidence. J Gerontol A Biol Sci Med Sci 2008; 63:

764e72

5. Cerejeira J, Firmino H, Vaz-Serra A, Mukaetova-

Ladinska EB. The neuroinflammatory hypothesis of

delirium. Acta Neuropathol 2010; 119: 737e54

6. Cunningham C, Maclullich AM. At the extreme end of the

psychoneuroimmunological spectrum: delirium as a

maladaptive sickness behaviour response. Brain Behav

Immun 2013; 28: 1e13

7. Sanders RD. Hypothesis for the pathophysiology of

delirium: role of baseline brain network connectivity and

changes in inhibitory tone. Med Hypotheses 2011; 77: 140e3

8. Tononi G. An information integration theory of con-

sciousness. BMC Neurosci 2004; 5: 42

9. Bastos AM, Usrey WM, Adams RA, Mangun GR, Fries P,

Friston KJ. Canonical microcircuits for predictive coding.

Neuron 2012; 76: 695e711

10. Numan T, Slooter AJC, van der Kooi AW, et al. Functional

connectivity and network analysis during hypoactive

delirium and recovery from anesthesia. Clin Neurophysiol

2017; 128: 914e24

11. van Dellen E, van der Kooi AW, Numan T, et al. Decreased

functional connectivity and disturbed directionality of

information flow in the electroencephalography of

intensive care unit patients with delirium after cardiac

surgery. Anesthesiology 2014; 121: 328e35

12 - Tanabe et al.

12. Murphy M, Riedner BA, Huber R, Massimini M, Ferrarelli F,

Tononi G. Source modeling sleep slow waves. Proc Natl

Acad Sci U S A 2009; 106: 1608e13

13. Murphy M, Bruno MA, Riedner BA, et al. Propofol anes-

thesia and sleep: a high-density EEG study. Sleep 2011; 34:

283e91

14. Ponten SC, Tewarie P, Slooter AJ, Stam CJ, van Dellen E.

Neural network modeling of EEG patterns in encepha-

lopathy. J Clin Neurophysiol 2013; 30: 545e52

15. Inouye SK, van Dyck CH, Alessi CA, Balkin S, Siegal AP,

Horwitz RI. Clarifying confusion: the confusion assess-

ment method. A new method for detection of delirium.

Ann Intern Med 1990; 113: 941e8

16. Marcantonio ER, Ngo LH, O’Connor M, et al. 3D-CAM:

derivation and validation of a 3-minute diagnostic inter-

view for CAM-defined delirium: a cross-sectional diag-

nostic test study. Ann Intern Med 2014; 161: 554e61

17. Ely EW, Inouye SK, Bernard GR, et al. Delirium in me-

chanically ventilated patients: validity and reliability of

the confusion assessment method for the intensive care

unit (CAM-ICU). JAMA 2001; 286: 2703e10

18. Trzepacz PT, Mittal D, Torres R, Kanary K, Norton J,

Jimerson N. Validation of the Delirium Rating Scale-

revised-98: comparison with the delirium rating scale

and the cognitive test for delirium. J Neuropsychiatry Clin

Neurosci 2001; 13: 229e42

19. Nir Y, Andrillon T, Marmelshtein A, et al. Selective

neuronal lapses precede human cognitive lapses

following sleep deprivation. Nat Med 2017; 23: 1474e80

20. Vinck M, Oostenveld R, van Wingerden M, Battaglia F,

Pennartz CM. An improved index of phase-

synchronization for electrophysiological data in the

presence of volume-conduction, noise and sample-size

bias. Neuroimage 2011; 55: 1548e65

21. Smith SM, Nichols TE. Threshold-free cluster enhance-

ment: addressing problems of smoothing, threshold

dependence and localisation in cluster inference. Neuro-

image 2009; 44: 83e98

22. Neter J, Kutner MH, Nachtsheim CJ, Wasserman W.

Applied linear statistical models. Chicago, IL: McGraw-Hill;

1996

23. Andersson JLR, Sotiropoulos SN. An integrated approach

to correction for off-resonance effects and subject move-

ment in diffusion MR imaging. Neuroimage 2016; 125:

1063e78

24. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in

functional and structural MR image analysis and imple-

mentation as FSL. Neuroimage 2004; 23: S208e19

25. Avants BB, Tustison NJ, Song G, Cook PA, Klein A, Gee JC.

A reproducible evaluation of ANTs similarity metric per-

formance in brain image registration. Neuroimage 2011; 54:

2033e44

26. Schwarz CG, Gunter JL, Ward CP, et al. The Mayo clinic

adult life span template: better quantification across the

life span. Alzheimers Dement 2017; 13: P93e4

27. Cavallari M, Dai W, Guttmann CR, et al. Neural substrates

of vulnerability to postsurgical delirium as revealed by

presurgical diffusion MRI. Brain 2016; 139: 1282e94

28. Smith SM, Jenkinson M, Johansen-Berg H, et al. Tract-

based spatial statistics: voxelwise analysis of multi-

subject diffusion data. Neuroimage 2006; 31: 1487e505

29. Vyazovskiy VV, Olcese U, Hanlon EC, Nir Y, Cirelli C,

Tononi G. Local sleep in awake rats. Nature 2011; 472:

443e7

30. van den Boogaard M, Kox M, Quinn KL, et al. Biomarkers

associated with delirium in critically ill patients and their

relation with long-term subjective cognitive dysfunction;

indications for different pathways governing delirium in

inflamed and noninflamed patients. Crit Care 2011; 15:

R297

31. Sala-Llonch R, Bartres-Faz D, Junque C. Reorganization of

brain networks in aging: a review of functional connec-

tivity studies. Front Psychol 2015; 6: 663

32. Babiloni C, Binetti G, Cassetta E, et al. Sources of cortical

rhythms change as a function of cognitive impairment in

pathological aging: a multicenter study. Clin Neurophysiol

2006; 117: 252e68

33. Babiloni C, Del Percio C, Lizio R, et al. Abnormalities of

resting state cortical EEG rhythms in subjects with mild

cognitive impairment due to Alzheimer’s and Lewy body

diseases. J Alzheimers Dis 2018; 62: 247e68

34. Gianotti LR, Kunig G, Lehmann D, et al. Correlation be-

tween disease severity and brain electric LORETA to-

mography in Alzheimer’s disease. Clin Neurophysiol 2007;

118: 186e96

35. Stam CJ, de Haan W, Daffertshofer A, et al. Graph theo-

retical analysis of magnetoencephalographic functional

connectivity in Alzheimer’s disease. Brain 2009; 132:

213e24

36. Engels MM, Stam CJ, van der Flier WM, Scheltens P, de

Waal H, van Straaten EC. Declining functional connec-

tivity and changing hub locations in Alzheimer’s disease:

an EEG study. BMC Neurol 2015; 15: 145

37. Cape E, Hall RJ, van Munster BC, et al. Cerebrospinal fluid

markers of neuroinflammation in delirium: a role for

interleukin-1beta in delirium after hip fracture.

J Psychosom Res 2014; 77: 219e25

38. Nakamura A, Cuesta P, Fernandez A, et al. Electromag-

netic signatures of the preclinical and prodromal stages of

Alzheimer’s disease. Brain 2018; 141: 1470e85

39. Reinhart RMG, Nguyen JA. Working memory revived in

older adults by synchronizing rhythmic brain circuits. Nat

Neurosci 2019; 22: 820e7

Handling editor: Michael Avidan